Influence of inner-core symmetry on tropical cyclone rapid intensification and its forecasting by a machine learning ensemble model

Abstract

This study proposed a novel quantitative index, the Symmetric Ratio, derived from satellite observations to depict Tropical Cyclone (TC) inner-core symmetry. This index is found to be significantly influential in TC Rapid Intensification (RI). We applied four machine learning (ML) models—Decision Tree, Random Forest, Light Gradient Boosting Machine, and Adaptive Boosting to forecast TC RI in the Northwestern Pacific (WNP) and North Atlantic (NA) basins from 2005 to 2023, with lead times of 12 and 24 hours. An ensemble model integrated these ML models to further enhance prediction accuracy. Model training used TC best track and reanalysis data from 2005 to 2020, with validation from 2021 to 2022. Independent forecasting tests from 2016 to 2023 applied real-time TC track data from the Automated Tropical Cyclone Forecasting system and environmental data from the Global Forecast System. Compared with the best deterministic model with the detection probability (POD) of 21 % and false alarm rate (FAR) of 50 % for 24-h RI forecasts in the NA basin during 2016–2020, our ensemble model demonstrated significant improvements, achieving a POD of 0.27 and an FAR of 0.18 for the same period. For 2021–2023, the ensemble model obtained POD values of 0.24 and 0.41, and FAR values of 0.33 and 0.45 for 24-h predictions in the NA and WNP basins, respectively. Key predictors identified include maximum wind speed tendency, vertical wind shear, potential intensity, and Symmetric Ratio. These findings advance our understanding of TC RI mechanisms and improve prediction accuracy.